Shogo Ando

Hello everyone.

This series will explain five tips to improve speech recognition accuracy. The third installment focuses on "How to properly use and configure microphones and apps."
In the previous article, we explained how to choose a microphone, but even if you purchase a microphone suitable for speech recognition, it will be useless if you use it incorrectly. It is also important to correctly use the microphone and set up the application that uses the microphone (for example, Zoom).

This time, we will explain the third tip for improving speech recognition accuracy: the importance of using the device properly.

Click here for the previous article ▼

How to Choose the Right Microphone - Speech recognition rate can change depending on the microphone!

Use the microphone correctly

First of all, it's very important to use the microphone correctly.
There are many different types of microphones, and each type has its own usage and precautions, so be sure to check the manual and use it accordingly.

We will explain using a headset microphone as an example.

The leftmost side of the illustration is the ideal way to use it.

Here are three examples of incorrect usage:

• Third from the left
  This is the case when the microphone is not close to the mouth. In this case, it cannot pick up much sound, so speech recognition is not good, but it is easy to notice mistakes and it is not a big problem in many cases.
• Third from the left
  This is the case when you hold the headset in your hand without wearing it. In this case, the microphone is near your mouth, but there is a possibility that sound will come from a different direction than the directional microphone.Although there is sound, the quality is low, which reduces the accuracy of voice recognition.Please be careful as this may occur.
• First from the right
  This is when the sponge at the tip of the microphone (called a windshield or windscreen) has come off. If the microphone is directly in front of your mouth and your breath hits it directly, or if there is wind around, you need to be careful not to blow the wind directly onto the microphone, which can result in a crackling sound. Some microphones are designed to keep the wind out even without a windshield, and in those cases, a windshield may not be included.

Select the correct input device in your OS or app settings

Next, we will explain how to correctly select the microphone you want to use in your OS or app.

It may seem obvious that you can't use a microphone unless you select it in the settings, but it's easy to make a careless mistake. In my experience, I was using a headset during an online meeting, but the built-in microphone of the webcam was actually set, and the headset's microphone wasn't being used at all. The webcam picked up the audio well enough, so the meeting wasn't affected, but because the webcam was far from my mouth, there was likely a lot of noise in the audio. I recommend always checking your settings to avoid poor sound quality due to the use of an unintended device.

For example, in the settings screen for the Windows version of Zoom below, you can select "マイク" and set the manufacturer of your microphone in the "オーディオ" drop-down menu.

Zoom's microphone settings also have the option to "システムと同じ". If you select this, the input device selected in Windows Sound Settings will be used.

Please note that there are different types of microphone settings depending on the application, such as "アプリケーションで独自に選択するもの", "Windowsの設定を使うもの" or "どちらかを選べるもの（Zoomはこれ）".

Set the microphone volume appropriately

Next, we will explain how to set the microphone volume.

Just like the microphone settings mentioned above, there are options such as "アプリケーションで独自に設定するもの" or "Windowsの設定を使うもの". Also, recently there are applications that have a function to automatically adjust the volume, so if such a setting is available, you can use it.

When adjusting the tone, there are two points to keep in mind: "Is the sound too quiet?" and "Is the sound distorted?"

The rightmost waveform in the image above is one where the volume is too loud, causing distortion. Distortion changes the physical characteristics of the sound, lowering the rate of speech recognition, so it's something you want to avoid as much as possible. The leftmost waveform is one where the volume is too quiet. If the volume is too quiet for humans to hear, the speech recognition engine won't be able to detect it, so this is also something you want to avoid. The volume of a voice fluctuates, so the appropriate volume is one where it's okay for it to get louder or quieter to a certain extent.

To adjust the volume, it is best to adjust it so that the meter does not swing out of control even when you speak fairly loudly, but still swings to a certain extent even when you speak quietly.

This table summarizes what happens if the volume is not appropriate. Please use it as a guide when making adjustments.

If it's too big	If it's too small
・If the sound is distorted, it will lead to a decrease in speech recognition accuracy. - Easily picks up surrounding noise	・Speech may not be detected, or the beginning and end of sentences may be omitted, leading to a decrease in speech recognition accuracy.

Maintain a consistent distance between the microphone and your mouth and speak at a constant volume

Even if you adjust the volume, if the distance between the microphone and your mouth changes significantly, the volume of your voice will change, so it is recommended that you keep the distance between your mouth and the microphone as constant as possible. Also, if the volume of your voice changes, it may cause distortion or sound too quiet. Please be careful, as the volume tends to decrease at the end of each sentence.

Pay attention to hardware connections and switches

Be careful of basic mistakes, such as "マイクを接続していなかった", "マイクのハードウェアスイッチでOFFになっていた" or "ソフトウェアでミュートになっていた".

On Windows, be careful of "排他モード"

There is "排他モード" setting in the microphone settings of Windows. If this is turned off, one microphone can be shared by multiple applications. However, if one application changes the microphone volume or other settings, it may affect other applications.

Turning it on will eliminate this effect, but since one microphone can only be used by one application, there is a possibility that recording will not be successful if an application other than the one you are trying to use speech recognition for monopolizes the microphone. This setting has its advantages and disadvantages, so you should consider it carefully before setting it.

Finally, listen to it with your own ears

Up until now, we have explained "the correct usage and configuration methods for microphones and apps", but we ultimately recommend that you record and confirm by listening with your own ears.

Even I sometimes think I'm properly inputting data into the microphone, but when I listen to the recorded audio later, I find that it sounds different from what I expected. Just to be sure, it's a good idea to record your voice and check whether the volume is sufficient, the sound quality, and whether there are any strange noises. If the volume and sound quality are easy for humans to hear, speech recognition should be easier.

If you use the speech recognition system while keeping in mind the points introduced here, you can expect to see improved speech recognition accuracy, so please give it a try.
And next time, we will explain about "using in a low-noise environment".

The accuracy of speech recognition (speech recognition rate) can be improved by the way you speak. What are the tips for how to speak?

How to Choose the Right Microphone - Speech recognition rate can change depending on the microphone!

Techniques to improve recognition accuracy even in noisy environments, the enemy of voice recognition

Is the Internet to blame for voice recognition not working? How to create a communication environment that allows voice recognition to work smoothly